mbox series

[v2,00/10] Fix up icc clock rate calculation on some platforms

Message ID 20230726-topic-icc_coeff-v2-0-8c91c6c76076@linaro.org
Headers show
Series Fix up icc clock rate calculation on some platforms | expand

Message

Konrad Dybcio July 31, 2023, 10:52 a.m. UTC 
Certain platforms require that some buses (or individual nodes) make
some additional changes to the clock rate formula, throwing in some
magic, Qualcomm-defined coefficients, to account for "inefficiencies".

Add the framework for it and utilize it on a couple SoCs.

Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
---
Changes in v2:
- Use the (arguably less favourable but necessary for precission) 100/x
  instead of x/100 for ib coefficient, update values in consequent
  patches to reflect that
- Rename "_percent" to "_coeff" because of /\
- Add the necessary code to support per-node clocks
- Add the necessary code to support per-node coefficients
- Hook up the CPUSS<->GNoC clock on QCM2290
- Update EBI node on QCM2290
- Link to v1: https://lore.kernel.org/r/20230726-topic-icc_coeff-v1-0-31616960818c@linaro.org

---
Konrad Dybcio (10):
      interconnect: qcom: icc-rpm: Add AB/IB calculations coefficients
      interconnect: qcom: icc-rpm: Separate out clock rate calulcations
      interconnect: qcom: icc-rpm: Let nodes drive their own bus clock
      interconnect: qcom: icc-rpm: Check for node-specific rate coefficients
      interconnect: qcom: qcm2290: Hook up MAS_APPS_PROC's bus clock
      interconnect: qcom: qcm2290: Set AB coefficients
      interconnect: qcom: qcm2290: Update EBI channel configuration
      interconnect: qcom: sdm660: Set AB/IB coefficients
      interconnect: qcom: msm8996: Set AB/IB coefficients
      clk: qcom: smd-rpm: Move CPUSS_GNoC clock to interconnect

 drivers/clk/qcom/clk-smd-rpm.c             | 16 ++++--
 drivers/interconnect/qcom/icc-rpm-clocks.c |  6 ++
 drivers/interconnect/qcom/icc-rpm.c        | 92 ++++++++++++++++++++++++------
 drivers/interconnect/qcom/icc-rpm.h        | 15 +++++
 drivers/interconnect/qcom/msm8996.c        |  8 ++-
 drivers/interconnect/qcom/qcm2290.c        |  9 ++-
 drivers/interconnect/qcom/sdm660.c         |  4 ++
 7 files changed, 124 insertions(+), 26 deletions(-)
---
base-commit: ec89391563792edd11d138a853901bce76d11f44
change-id: 20230726-topic-icc_coeff-b053d5409b9f

Best regards,

Comments

Stephan Gerhold Aug. 1, 2023, 11:01 a.m. UTC | #1 
On Mon, Jul 31, 2023 at 12:52:20PM +0200, Konrad Dybcio wrote:
> Some nodes may have different coefficients than the general values for
> bus they're attached to. Check for that and use them if present.
> 
> Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
> ---
>  drivers/interconnect/qcom/icc-rpm.c | 10 +++++++---
>  drivers/interconnect/qcom/icc-rpm.h |  6 ++++++
>  2 files changed, 13 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/interconnect/qcom/icc-rpm.c b/drivers/interconnect/qcom/icc-rpm.c
> index f0e575c95b49..91eb428385f6 100644
> --- a/drivers/interconnect/qcom/icc-rpm.c
> +++ b/drivers/interconnect/qcom/icc-rpm.c
> @@ -300,11 +300,15 @@ static u64 qcom_icc_calc_rate(struct qcom_icc_provider *qp, struct qcom_icc_node
>  	else
>  		agg_avg_rate = qn->sum_avg[ctx];
>  
> -	/* Check if the node has a specific coefficient first*/
> -	if (qp->ab_coeff)
> +	/* Check if the node has a specific coefficient first */
> +	if (qn->ab_coeff)
> +		agg_avg_rate = mult_frac(qn->ab_coeff, agg_avg_rate, 100);
> +	else if (qp->ab_coeff)
>  		agg_avg_rate = mult_frac(qp->ab_coeff, agg_avg_rate, 100);
>  
> -	if (qp->ib_coeff)
> +	if (qn->ab_coeff)
> +		agg_peak_rate = mult_frac(100, qn->max_peak[ctx], qn->ib_coeff);
> +	else if (qp->ib_coeff)
>  		agg_peak_rate = mult_frac(100, qn->max_peak[ctx], qp->ib_coeff);
>  	else
>  		agg_peak_rate = qn->max_peak[ctx];

Code/data size and likely performance would be slightly better if you
only add ab_coeff/ib_coeff to the node and not the provider. This is
slightly inconvenient because you need to duplicate the same value on
a lot of nodes, but the per-node memory is reserved anyway. You might
as well use I would say.

> diff --git a/drivers/interconnect/qcom/icc-rpm.h b/drivers/interconnect/qcom/icc-rpm.h
> index 835b83cfb548..1a26a7b82166 100644
> --- a/drivers/interconnect/qcom/icc-rpm.h
> +++ b/drivers/interconnect/qcom/icc-rpm.h
> @@ -103,6 +103,9 @@ struct qcom_icc_qos {
>   * @mas_rpm_id:	RPM id for devices that are bus masters
>   * @slv_rpm_id:	RPM id for devices that are bus slaves
>   * @qos: NoC QoS setting parameters
> + * @ab_coeff: a percentage-based coefficient for compensating the AB calculations
> + * @ib_coeff: an inverse-percentage-based coefficient for compensating the IB calculations
> + * @bus_clk_rate: a pointer to an array containing bus clock rates in Hz
>   */
>  struct qcom_icc_node {
>  	unsigned char *name;
> @@ -117,6 +120,9 @@ struct qcom_icc_node {
>  	int mas_rpm_id;
>  	int slv_rpm_id;
>  	struct qcom_icc_qos qos;
> +	u16 ab_coeff;
> +	u16 ib_coeff;
> +	u32 *bus_clk_rate;

bus_clk_rate should be in previous patch :)

Thanks,
Stephan
Georgi Djakov Aug. 4, 2023, 4:31 p.m. UTC | #2 
Hi Konrad,

On 31.07.23 13:52, Konrad Dybcio wrote:
> Certain platforms require that some buses (or individual nodes) make
> some additional changes to the clock rate formula, throwing in some
> magic, Qualcomm-defined coefficients, to account for "inefficiencies".

Maybe some links to the downstream code would help to better check and
understand this. Adding also Odelu in case he has any comments on the
patches.

Thanks,
Georgi

> Add the framework for it and utilize it on a couple SoCs.
> 
> Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
> ---
> Changes in v2:
> - Use the (arguably less favourable but necessary for precission) 100/x
>    instead of x/100 for ib coefficient, update values in consequent
>    patches to reflect that
> - Rename "_percent" to "_coeff" because of /\
> - Add the necessary code to support per-node clocks
> - Add the necessary code to support per-node coefficients
> - Hook up the CPUSS<->GNoC clock on QCM2290
> - Update EBI node on QCM2290
> - Link to v1: https://lore.kernel.org/r/20230726-topic-icc_coeff-v1-0-31616960818c@linaro.org
> 
> ---
> Konrad Dybcio (10):
>        interconnect: qcom: icc-rpm: Add AB/IB calculations coefficients
>        interconnect: qcom: icc-rpm: Separate out clock rate calulcations
>        interconnect: qcom: icc-rpm: Let nodes drive their own bus clock
>        interconnect: qcom: icc-rpm: Check for node-specific rate coefficients
>        interconnect: qcom: qcm2290: Hook up MAS_APPS_PROC's bus clock
>        interconnect: qcom: qcm2290: Set AB coefficients
>        interconnect: qcom: qcm2290: Update EBI channel configuration
>        interconnect: qcom: sdm660: Set AB/IB coefficients
>        interconnect: qcom: msm8996: Set AB/IB coefficients
>        clk: qcom: smd-rpm: Move CPUSS_GNoC clock to interconnect
> 
>   drivers/clk/qcom/clk-smd-rpm.c             | 16 ++++--
>   drivers/interconnect/qcom/icc-rpm-clocks.c |  6 ++
>   drivers/interconnect/qcom/icc-rpm.c        | 92 ++++++++++++++++++++++++------
>   drivers/interconnect/qcom/icc-rpm.h        | 15 +++++
>   drivers/interconnect/qcom/msm8996.c        |  8 ++-
>   drivers/interconnect/qcom/qcm2290.c        |  9 ++-
>   drivers/interconnect/qcom/sdm660.c         |  4 ++
>   7 files changed, 124 insertions(+), 26 deletions(-)
> ---
> base-commit: ec89391563792edd11d138a853901bce76d11f44
> change-id: 20230726-topic-icc_coeff-b053d5409b9f
> 
> Best regards,